3D Measurement & Virtual Reconstruction of Ancient Lost Worlds of Europe

This 3D surface acquisition technique can readily be applied to archaeological field work. The on-site acquisition procedure consists of recording an image sequence of the scene that one desires to reconstruct. To allow the algorithms to yield good results viewpoint changes between consecutive images should not exceed 5 to 10 degrees.

A key feature of the approach - coined `shape-from-video' - is that the necessary camera calibration is carried out automatically, from the same image data that is used for the generation of the 3D reconstructions. Hence, the sole input that this approach needs are images. No knowledge is necessary about the camera settings or about the relative positions from where images are taken.

A first example is one of the Dionysos statues found in Sagalassos. In antiquity, this 2m high statue was placed on the monumental fountain (nymphaeum) situated at the upper market square of Sagalassos. The statue is now located in the garden of the museum in Burdur. Producing the 3D reconstruction only required the recording of a 1-minute video. Bringing in special equipment such as a laser range scanner would not only have been logistically more demanding, it would also have required a special authorisation. In Figure 1 different steps of the reconstruction process are illustrated.

The 3D model was obtained from a single depth map. A more complete and accurate model could be obtained by combining multiple depth maps. A smoother look for the shaded model could be obtained by filtering the 3D mesh in accordance with the standard deviations that are obtained as a by-product of the depth computation. This is not so important when the model is texture mapped with the original images. For many visualisation purposes, this result will suffice.

Figure 1: 3D reconstruction of Dionysos. (a) one of the original video frames, (b) corresponding depth map, (c) shaded view of the 3D reconstruction, (d) view of the textured 3D model.

A second example is shown in Figure 2. It is a Medusa head which is located on the entablature of the fountain building. The head itself is about 30cm large. The 3D model was obtained from a short video sequence. In this case a single depth map was also used to reconstruct the 3D model. Notice that realistic views can be rendered from viewpoints that are very different from the original viewpoint.

Figure: 3D reconstruction of a Medusa head. (a) one of the original video frames, (b) corresponding depth map, (c) and (d) two views of the 3D model.

An important advantage of our approach is its ease-of-use. Taking images is part of regular, archaeological practice. The ability to turn normal photographs into 3D models literally adds a new dimension to archaeological records. Compared to non-image based techniques there is the important advantage that surface texture is directly extracted from the images. This
adds to the realism and authenticity of the reconstructions. A disadvantage of the approach (and more in general of most image-based approaches) is that it does not directly capture the photometric properties of an object, but
only their combination with lighting. It is therefore not possible to render the 3D model under different lighting. Recent work (e.g. by Debevec et al.,USC) does allow to change the lighting for archaeological artefacts. There an image-based approach is also used. But in that case, use is made of a controlled lighting set-up during acquisition. It remains to be investigated what can be done in uncontrolled lighting environments like ours.

As many of the bigger archaeological objects and scenes do not lend themselves to be captured in one continuous image sequence, a new tool (‘the Layer Matcher’) has been developed that allows for the merging of different 3D reconstructed parts of the same scene in one single framework. This is accomplished by querying the user for some corresponding points between the different reconstructions. This tool demonstrated to be especially useful for the generation of 3D data for stratigraphical layers. It seems to be difficult to capture an entire layer with a single image sequence. The next figure shows how two 3D reconstructions, originating from separate image sequences, are merged together into a single framework by indicating some corresponding points between both reconstructions.

Figure: Top: Two 3D reconstructed parts of the same stratigraphic layer originating from two separate image sequences. Bottom: The merged result of both reconstruction into a single framework after indication of some corresponding points in both reconstructions (red points in top figures).

Another feature of the ‘Layer Matcher’ is the ability to merge 3D reconstructions of consecutive stratigrahical layers as the archaeologists progress and uncover new stratigraphical layers, thereby destroying the old layers. The tool makes it possible to keep a 3D record of all layers in time, a valuable advantage due to the destructive nature of the excavation of stratigraphic layers. The next figure demonstrates how two consecutive layers (in time) can be merged together to give an impression of the excavation progress.

Figure: Top: Two reconstructions of the same area at different times. The left layer is destroyed to uncover the right layer. Bottom: The integration of both reconstructions. The top layer appears to be floating above the bottom layer. This gives an impression of the excavation activity between two time instances.